What happened
GitHub has rolled out custom agents for Copilot CLI, enabling engineering teams to encode their specific stack configurations and internal workflows directly into the terminal tool. Rather than issuing isolated, ad-hoc prompts, developers can now build standardized, repeatable processes that reflect how their organization actually operates. Separately, Vercel's AI Gateway production index for June 2026 reveals that DeepSeek has entered serious competition for raw token volume, while Anthropic continues to command the largest share of actual spend among the models transacting through the platform.
Why it matters for your business
For engineering leaders, the GitHub Copilot CLI update addresses a persistent gap: AI assistance has largely been ad-hoc, making it hard to enforce consistency or conduct post-hoc review of AI-assisted work. Custom agents introduce auditability and repeatability, which are prerequisites for regulated industries or teams scaling beyond a handful of engineers. On the model-market side, the Vercel data signals a bifurcating competitive landscape — DeepSeek's volume gains suggest cost-sensitive workloads are migrating toward lower-price options, but Anthropic's spending dominance indicates enterprises are still paying a premium for higher-stakes tasks. Operations and procurement leaders evaluating AI contracts should factor in this cost-versus-criticality split when negotiating model usage agreements.
What to watch next
The adoption curve for GitHub Copilot CLI agents will be a bellwether for whether AI-assisted development matures from individual productivity tool to team-level infrastructure. On the model side, if DeepSeek continues closing the gap on spend — not just volume — it could pressure Anthropic and others to revisit pricing tiers for high-throughput workloads. Vercel's ongoing production index will be worth monitoring quarter-over-quarter as a real-world proxy for enterprise model preference.